At John Deere, we use machine learning and computer vision (including stereo vision) for challenging outdoor applications such as obstacle detection, vision-based guidance and weed management, among many others. The quality of the images our systems obtain, and the accuracy of the depth information produced by our stereo cameras, significantly impact the performance of the overall solutions. In this talk, we’ll share some of the challenges we’ve faced in developing image quality improvement and stereo vision algorithms. Many of the techniques found in academic research and prior work cannot be easily implemented in real-time applications at the edge and are difficult to scale for applications with varying performance and cost requirements. We’ll highlight some of the alternative techniques we’ve developed to provide optimized image-quality and stereo vision implementations that meet the requirements of John Deere’s product range. Stated another way, we’ll share how we do more with less.